muji_moe

Lightweight desktop voice chatbot developed by Reecho.ai, using Live2D Cubism SDK and C++. Reecho.ai is a group that focuses on multilanguage speech synthesis (Chinese, English...), and this project is a part of our research on voice chatbots. In this project, we use Live2D Cubism SDK to create a 2D character to interact with users. Customized OpenAI API is supported in this project to generate chat content and synthesize speech using Reecho.ai's speech synthesis engine. Furthermore, switching to RTC/WS mode can get the low-latency voice chat experience (~800 ms) but cost more tokens. Live2D MUJI_MOE model is created by XWdit and is for non-commercial use only.

Roadmap

• Basic UI
• Chatbot
• Customized OpenAI API
• Voice chat (TTS Mode)
• Voice chat (RTC/WS Mode)
• Windows support
• More input examples

Video Demo

muji_moe.mp4

Build Requirements

Put following libraries in lib/ directory:

• Live2D Cubism Core SDK For Native(Frameworks) [5-r.1]
• asio [1.30.2]
• cpr [1.11.0]
• glew [2.1.0]
• glfw [3.4.0]
• libsoundio [1.1.0]
• jsoncpp
• stb_image

Build and Run

cd mac && mkdir build && cd build
cmake .. && make
cd build/bin/ && ./muji_moe

License

• MUJI_MOE Live2D Model (Resources/muji_moe_auto) is licensed under AGPL-3.0 License - see the LICENSE MUJI MOE.
• Live2D Cubism SDK is licensed under the Live2D Proprietary Software License Agreement.
• Live2D Model is licensed under the Live2D Proprietary Software License Agreement.
• Source code is licensed under the MIT License - see the LICENSE file for details.